VIABench: A Comprehensive Video Benchmark Collected from Blind Individuals for Visual Impairment Assistance
The benchmark uses first-person videos recorded or shared by blind people to test whether MLLMs can actually assist in visual impairment scenarios.
VIABench covers proactive reminders, visual question answering, and vision-guided interaction. It supports both real-time and offline evaluation. The authors report that current multimodal models still fall short, especially when they must anticipate navigation-critical events quickly. Code and data are planned for release. HF Daily Papers' note
VIABench covers proactive reminders, visual question answering, and vision-guided interaction. It supports both real-time and offline evaluation. The authors report that current multimodal models still fall short, especially when they must anticipate navigation-critical events quickly. Code and data are planned for release. HF Daily Papers' note
score 5